Two Discoveries, One Principle: Using a Two-Stage Bayesian Model to Explain a Dissociated Working Memory Distraction Effect

How working memory (WM) resists perceptual distraction with its limited capacity is a fundamental question to understand its mechanism. To address this question, we used a continuous recall paradigm to directly compare the distraction effect during encoding and the delay periods. Across Experiments 1-3, we observed a substantial distraction-related cost on mnemonic fidelity when distractors presented during the delay (delay-distraction condition), but not if they were introduced at encoding (encoding-distraction condition) or across both periods (full-distraction condition). However, the distraction cost revived when we increased the difficulty to distinguish distractors from targets (Experiments 4 and 4S) and when we changed distractors' relevant features during the delay (Experiment 5). We also found that the robust distraction cost in the delay-distraction condition did not occupy extra spatial resources (Experiments 6a and 6b). These results suggested a dissociated distraction effect, which could be related to the dynamic resource allocation across two WM periods. Here, we proposed a Bayesian model and considered the task relevance and visual uncertainty as two main factors that determine the resource allocation principle at two different stages. This model successfully captured the main findings across all behavioral experiments and performed better than other alternative models. Taken together, the current work advanced our understanding of the distraction resistance of WM under the framework of limited resource allocation.